SGS-Planner: A Skeleton-Guided Spatiotemporal Motion Planner for Flight in Constrained Space

This article proposes a skeleton-guided spatiotemporal motion planner (SGS-Planner) for safe and efficient aerial robot navigation in challenging constrained space. The planner takes advantage of a spatial-temporal hierarchical framework to generate a safer and more efficient trajectory rapidly. In spatial planning, different from existing approaches that either ignore the clearance between trajectories and obstacles or suffer from local minima, path smoothing is guided by a skeleton away from obstacles. Specifically, a fast sphere inflation-based skeleton extraction method is elaborately designed for high-clearance path searching. After that, we propose a skeleton-guided path smoothing approach to generate smooth and high-clearance paths, where the smoothing is formulated as unconstrained quadratic programs. In temporal planning, the feasible trajectory is generated along the optimized path by employing an optimal time allocation method and a parallel back-up strategy to guarantee the success of planning. We validate the performance of the proposed method in extensive challenging simulations and real-world environments. Comparative results show that our approach outperforms state-of-the-art methods in terms of planning success rate, average computational efficiency (5–24 times faster), and minimum clearance (12%–177% larger) in restricted space.

I. Introduction

Unmanned aerial vehicles (UAVs) with mobility and flexibility have been widely applied in various scenarios, such as exploration [1], inspection [2], search and rescue [3], and so on. Motion planning is one of the key techniques for completing the abovementioned missions. There are considerable motion planning methods developed for navigation in cluttered environments [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18]. However, in many circumstances, UAVs are desired to fly in constrained spaces, such as mines [19], caves [20], and so on. It is still a challenge to navigate in such environments for existing approaches.